I didn’t like the idea of using GPUs as general purpose processing units.
Ok, TBH I still don’t like it.

It’s hard to find somebody loving multithreaded programming but nowadays multicore CPUs are here and we have to adapt. The same applies to GPUs, as many hundreds processing units are stored inside a single GPU core , delivering up to a couple of TFLOPS of raw processing power, we have to adapt and think beyond polygons.

Some months ago I installed CUDA and did a few tests. It was a bit weird to compile CUDA programs with a special compiler and in general the setup process didn’t look clean. I decided to stop experimenting until something compatible with all GPUs would be available.

Now, as Compute Shaders and OpenCL are available, I decided to jump on the GPU computing bandwagon.  Here’s a brief report of my first steps in this new world. This post deals with configuring a GPU computing environment with an old nVidia GPU.

(continue reading…)